CASE STUDY
PRISM (Zusman Research)
TBI research database and analysis platform
THE SITUATION
A 3,473-patient cohort living across spreadsheets and PDFs
The Zusman Research group at the Prisma Institute (PRISM) had spent years collecting longitudinal data on traumatic brain injury outcomes. The clinical intake was rigorous; the infrastructure behind it was not. Records lived in a tangle of exports, spreadsheets, and legacy database snapshots.
Any real analysis required a week of preparation work before a single question could be asked. The team needed a foundation they could build research on, not a mountain of cleanup.
BEFORE
A week of cleanup before any real question could be asked.
AFTER
The cohort is queryable. Researchers run their own analyses without blocking on engineering.
WHAT WE BUILT
A research database the clinical team co-designed
We designed the schema with the researchers, not for them. Normalized patient, visit, and outcome tables modeled on how the clinical team actually reasons about the cohort. Postgres as the source of truth; Snowflake for the heavier analytical cuts.
ETL pipelines reconcile the legacy exports against the new model, surfacing conflicts to the clinical lead rather than silently picking a winner.
Clinical-first schema
Entities and relationships reviewed line by line with the lead researcher before a single migration ran.
Reconciled ingestion
Historical spreadsheets and database snapshots flow through a pipeline that flags discrepancies instead of hiding them.
Analytical surface
Snowflake warehouse backs the analysis notebooks; the clinical team runs their own queries without asking engineering.
THE OUTCOMES
Used every week by the people it was built for
3,473
patient cohort, fully migrated
Weekly
use by the clinical team since launch
2
manuscripts in progress on data this platform surfaced
STACK
- Postgres
- Snowflake
- Python
- dbt
- Airflow
- Typescript
